Unsupervised classification of plethysmography signals with advanced visual representations

Abstract

Ventilation is a simple physiological function that ensures the vital supply of oxygen and the elimination of CO₂. The recording of the airflow through the nostrils of a mouse over time makes it possible to calculate the position of critical points, based on the shape of the signals, to compute the respiratory frequency and the volume of air exchanged. These descriptors only account for a part of the dynamics of respiratory exchanges. In this work we present a new algorithm that directly compares the shapes of signals and considers meaningful information about the breathing dynamics omitted by the previous descriptors. The algorithm leads to a new classification of inspiration and expiration, which reveals that mice respond and adapt differently to inhibition of cholinesterases, enzymes targeted by nerve gas, pesticide, or drug intoxication.

Keywords: breathing; clustering; dynamic time warping (DTW); machine learning; respiration.

FIGURE 1

(A): Illustration of a double-chamber plethysmograph. The term dpt stands for differential pressure transducer which measures the pressure in each compartment, the pressure then being converted to flow. (B): Nasal airflow (top) and lung volume (bottom). During inspiration, airflow is positive (grey) and during expiration, airflow is negative (pink).

FIGURE 2

(A): Step 1, Detection of the respiratory cycles and extraction of the inspiration/expiration sequences. (B): Step 2, Computation of the reference sequences. The C ⁽ⁱ⁾ denote the clusters and r ⁽ⁱ⁾, the reference sequences. (C): Step 3, Characterization and symbolization of a recordings.

FIGURE 3

(A): Undesirable sequence distortions. The grey line is the original signal and the blue line is the distorted signal. (B): Difference between Euclidean alignment and DTW alignment. The compared sequences are in blue, and the orange lines represent the point-wise matching between the two sequences in the Euclidean case and the DTW case. (C): Representation of a barycenter computed through our method BS-DBA. The learned barycenter is in red, and the observations are in black.

FIGURE 4

Respiratory cycle map displays with nasal airflow (mL.s ⁻¹) on the left and nasal volume (mL) on the right. Positive flow corresponds to inspiration and negative flow corresponds to expiration.

FIGURE 5

(A): Respiratory Cycle map (RC map) built-up process. (B): Respiratory RC maps: All RC maps are truncated at the threshold value of 20%. RC maps are grouped by genotype: WT, PRIMA, AChE1iRR, ColQ. For each genotype, the two left columns and the two right columns gathered RC maps respectively before and after physostigmine injection. Numbers on RC maps correspond to the mouse id. The bottom line corresponds to the average RC maps observed per genotype before and after drug injection. (C): Average reference sequence polar plots: Polar plots are grouped by genotype. Inspirations are on the top, and expirations are on the bottom. The values on each angular axis correspond to the average percentage of time assigned to the associated reference sequence. The blue polygon corresponds to the values observed before injection, and the red polygon corresponds to the values observed after the injection.

FIGURE 6

(A): Respiratory bar codes built-up process.(B): Respiratory bar codes are gathered by genotype: (top,left): WT, (bottom, left): PRIMA, (top,right): AChE1iRR, (bottom, right): ColQ. Numbers to the left of bar codes correspond to the mouse id. For each genotype, the left section corresponds to barcodes before drug injection and the right section to bar codes after injection. Grey areas in bar codes like mouse PRIMA-2 correspond to unpredictable cycles. Some experiments were shorter than others resulting in shorter bar codes.

FIGURE 7

Multiple testing scheme with a false discovery rate (FDR) correction of 5%, performing a Mann-Whitney U test for each type of respiratory cycle. A cell is colored black if the unit null hypothesis is rejected after FDR correction and includes the corrected p-value of the associated unit test. (A): Statistical tests comparing the distribution of respiratory cycles of control (WT) and AChE-deficient (PRIMA, AChE1iRR, COLQ) mice before drug injection. (B): Statistical tests comparing the distribution of respiratory cycles before and after drug injection for each genotype.

FIGURE 8

Examples of typical respiratory behaviors. For each panel, the left column represents the referent cycle, and the right column is an extract from a recording of up to 5 s where the reference cycle is repeated continuously. Charts with a blue background are expressed in nasal airflow, and charts with a yellow background are expressed in nasal volume. (A): Referent cycle B0. (B): Referent cycle B4. Inspiration, expiration, and end-inspiratory pause (EIP) duration are illustrated. (C): Referent cycle C0. (D): Referent cycle D0.

FIGURE 9

Box plots of the respiratory cycle descriptors: inspiration/expiration time and inspiration/expiration volume. Each box plot represents a referent sequence. A box represents the first quartile (Q1), median, and third quartile (Q3). The lower whisker corresponds to the minimum value observed, and the upper whisker is above the third quartile by 1.5 interquartile range (IQR: Q3-Q1).